Multi-Objective Evolutionary Algorithms for Multivariate Optimal Allocation

• Activity Number: 484
• Type: Contributed
• Date/Time: Wednesday, August 4, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Charles D. Day*+
• Companies: IRS
• Keywords: Sample Design ; Genetic Algorithm

Abstract

Typically, statisticians approach multivariate optimal allocation in one of two ways. Either they set target CVs for the most important variables and minimize sample size with those targets as constraints, or they minimize a function of the CVs that implies some preference between variables constrained by a fixed total sample size. This paper presents an alternative to those approaches. A set of non-dominated trade-off solutions is produced using a multi-objective evolutionary algorithm. A solution is non-dominated if no other solution exists that is better on one or more objectives and at least as good on all others. By examining sets of non-dominated solutions, survey designers can see the trade-offs being made between them, rather than having to choose an arbitrary preference function prior to beginning sample allocation.